json 2.12.2 → 2.15.0

data/ext/json/ext/generator/generator.c

@@ -5,10 +5,16 @@
 #include <math.h>
 #include <ctype.h>
 
-#include "simd.h"
+#include "../simd/simd.h"
 
 /* ruby api and some helpers */
 
+enum duplicate_key_action {
+    JSON_DEPRECATED = 0,
+    JSON_IGNORE,
+    JSON_RAISE,
+};
+
 typedef struct JSON_Generator_StateStruct {
     VALUE indent;
     VALUE space;
@@ -21,6 +27,9 @@ typedef struct JSON_Generator_StateStruct {
     long depth;
     long buffer_initial_length;
 
+    enum duplicate_key_action on_duplicate_key;
+
+    bool as_json_single_arg;
     bool allow_nan;
     bool ascii_only;
     bool script_safe;
@@ -31,10 +40,10 @@ typedef struct JSON_Generator_StateStruct {
 #define RB_UNLIKELY(cond) (cond)
 #endif
 
-static VALUE mJSON, cState, cFragment, mString_Extend, eGeneratorError, eNestingError, Encoding_UTF_8;
+static VALUE mJSON, cState, cFragment, eGeneratorError, eNestingError, Encoding_UTF_8;
 
 static ID i_to_s, i_to_json, i_new, i_pack, i_unpack, i_create_id, i_extend, i_encode;
-static VALUE sym_indent, sym_space, sym_space_before, sym_object_nl, sym_array_nl, sym_max_nesting, sym_allow_nan,
+static VALUE sym_indent, sym_space, sym_space_before, sym_object_nl, sym_array_nl, sym_max_nesting, sym_allow_nan, sym_allow_duplicate_key,
              sym_ascii_only, sym_depth, sym_buffer_initial_length, sym_script_safe, sym_escape_slash, sym_strict, sym_as_json;
 
 
@@ -137,8 +146,8 @@ static inline FORCE_INLINE void search_flush(search_state *search)
 {
     // Do not remove this conditional without profiling, specifically escape-heavy text.
     // escape_UTF8_char_basic will advance search->ptr and search->cursor (effectively a search_flush).
-    // For back-to-back characters that need to be escaped, specifcally for the SIMD code paths, this method
-    // will be called just before calling escape_UTF8_char_basic. There will be no characers to append for the
+    // For back-to-back characters that need to be escaped, specifically for the SIMD code paths, this method
+    // will be called just before calling escape_UTF8_char_basic. There will be no characters to append for the
     // consecutive characters that need to be escaped. While the fbuffer_append is a no-op if
     // nothing needs to be flushed, we can save a few memory references with this conditional.
     if (search->ptr > search->cursor) {
@@ -304,28 +313,6 @@ static inline FORCE_INLINE unsigned char neon_next_match(search_state *search)
     return 1;
 }
 
-// See: https://community.arm.com/arm-community-blogs/b/servers-and-cloud-computing-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon
-static inline FORCE_INLINE uint64_t neon_match_mask(uint8x16_t matches)
-{
-    const uint8x8_t res = vshrn_n_u16(vreinterpretq_u16_u8(matches), 4);
-    const uint64_t mask = vget_lane_u64(vreinterpret_u64_u8(res), 0);
-    return mask & 0x8888888888888888ull;
-}
-
-static inline FORCE_INLINE uint64_t neon_rules_update(const char *ptr)
-{
-    uint8x16_t chunk = vld1q_u8((const unsigned char *)ptr);
-
-    // Trick: c < 32 || c == 34 can be factored as c ^ 2 < 33
-    // https://lemire.me/blog/2025/04/13/detect-control-characters-quotes-and-backslashes-efficiently-using-swar/
-    const uint8x16_t too_low_or_dbl_quote = vcltq_u8(veorq_u8(chunk, vdupq_n_u8(2)), vdupq_n_u8(33));
-
-    uint8x16_t has_backslash = vceqq_u8(chunk, vdupq_n_u8('\\'));
-    uint8x16_t needs_escape  = vorrq_u8(too_low_or_dbl_quote, has_backslash);
-
-    return neon_match_mask(needs_escape);
-}
-
 static inline unsigned char search_escape_basic_neon(search_state *search)
 {
     if (RB_UNLIKELY(search->has_matches)) {
@@ -333,7 +320,7 @@ static inline unsigned char search_escape_basic_neon(search_state *search)
         if (search->matches_mask > 0) {
             return neon_next_match(search);
         } else {
-            // neon_next_match will only advance search->ptr up to the last matching character. 
+            // neon_next_match will only advance search->ptr up to the last matching character.
             // Skip over any characters in the last chunk that occur after the last match.
             search->has_matches = false;
             search->ptr = search->chunk_end;
@@ -342,67 +329,61 @@ static inline unsigned char search_escape_basic_neon(search_state *search)
 
     /*
     * The code below implements an SIMD-based algorithm to determine if N bytes at a time
-    * need to be escaped. 
-    * 
+    * need to be escaped.
+    *
     * Assume the ptr = "Te\sting!" (the double quotes are included in the string)
-    * 
+    *
     * The explanation will be limited to the first 8 bytes of the string for simplicity. However
     * the vector insructions may work on larger vectors.
-    * 
+    *
     * First, we load three constants 'lower_bound', 'backslash' and 'dblquote" in vector registers.
-    * 
-    * lower_bound: [20 20 20 20 20 20 20 20] 
-    * backslash:   [5C 5C 5C 5C 5C 5C 5C 5C] 
-    * dblquote:    [22 22 22 22 22 22 22 22] 
-    * 
-    * Next we load the first chunk of the ptr: 
+    *
+    * lower_bound: [20 20 20 20 20 20 20 20]
+    * backslash:   [5C 5C 5C 5C 5C 5C 5C 5C]
+    * dblquote:    [22 22 22 22 22 22 22 22]
+    *
+    * Next we load the first chunk of the ptr:
     * [22 54 65 5C 73 74 69 6E] ("  T  e  \  s  t  i  n)
-    * 
+    *
     * First we check if any byte in chunk is less than 32 (0x20). This returns the following vector
     * as no bytes are less than 32 (0x20):
     * [0 0 0 0 0 0 0 0]
-    * 
+    *
     * Next, we check if any byte in chunk is equal to a backslash:
     * [0 0 0 FF 0 0 0 0]
-    * 
+    *
     * Finally we check if any byte in chunk is equal to a double quote:
-    * [FF 0 0 0 0 0 0 0] 
-    * 
+    * [FF 0 0 0 0 0 0 0]
+    *
     * Now we have three vectors where each byte indicates if the corresponding byte in chunk
     * needs to be escaped. We combine these vectors with a series of logical OR instructions.
     * This is the needs_escape vector and it is equal to:
-    * [FF 0 0 FF 0 0 0 0] 
-    * 
+    * [FF 0 0 FF 0 0 0 0]
+    *
     * Next we compute the bitwise AND between each byte and 0x1 and compute the horizontal sum of
     * the values in the vector. This computes how many bytes need to be escaped within this chunk.
-    * 
+    *
     * Finally we compute a mask that indicates which bytes need to be escaped. If the mask is 0 then,
     * no bytes need to be escaped and we can continue to the next chunk. If the mask is not 0 then we
     * have at least one byte that needs to be escaped.
     */
-    while (search->ptr + sizeof(uint8x16_t) <= search->end) {
-        uint64_t mask = neon_rules_update(search->ptr);
 
-        if (!mask) {
-            search->ptr += sizeof(uint8x16_t);
-            continue;
-        }
-        search->matches_mask = mask;
+    if (string_scan_simd_neon(&search->ptr, search->end, &search->matches_mask)) {
         search->has_matches = true;
         search->chunk_base = search->ptr;
         search->chunk_end = search->ptr + sizeof(uint8x16_t);
         return neon_next_match(search);
     }
 
-    // There are fewer than 16 bytes left. 
+    // There are fewer than 16 bytes left.
     unsigned long remaining = (search->end - search->ptr);
     if (remaining >= SIMD_MINIMUM_THRESHOLD) {
         char *s = copy_remaining_bytes(search, sizeof(uint8x16_t), remaining);
 
-        uint64_t mask = neon_rules_update(s);
+        uint64_t mask = compute_chunk_mask_neon(s);
 
         if (!mask) {
-            // Nothing to escape, ensure search_flush doesn't do anything by setting 
+            // Nothing to escape, ensure search_flush doesn't do anything by setting
             // search->cursor to search->ptr.
             fbuffer_consumed(search->buffer, remaining);
             search->ptr = search->end;
@@ -428,11 +409,6 @@ static inline unsigned char search_escape_basic_neon(search_state *search)
 
 #ifdef HAVE_SIMD_SSE2
 
-#define _mm_cmpge_epu8(a, b) _mm_cmpeq_epi8(_mm_max_epu8(a, b), a)
-#define _mm_cmple_epu8(a, b) _mm_cmpge_epu8(b, a)
-#define _mm_cmpgt_epu8(a, b) _mm_xor_si128(_mm_cmple_epu8(a, b), _mm_set1_epi8(-1))
-#define _mm_cmplt_epu8(a, b) _mm_cmpgt_epu8(b, a)
-
 static inline FORCE_INLINE unsigned char sse2_next_match(search_state *search)
 {
     int mask = search->matches_mask;
@@ -457,18 +433,6 @@ static inline FORCE_INLINE unsigned char sse2_next_match(search_state *search)
 #define TARGET_SSE2
 #endif
 
-static inline TARGET_SSE2 FORCE_INLINE int sse2_update(const char *ptr)
-{
-    __m128i chunk         = _mm_loadu_si128((__m128i const*)ptr);
-
-    // Trick: c < 32 || c == 34 can be factored as c ^ 2 < 33
-    // https://lemire.me/blog/2025/04/13/detect-control-characters-quotes-and-backslashes-efficiently-using-swar/
-    __m128i too_low_or_dbl_quote = _mm_cmplt_epu8(_mm_xor_si128(chunk, _mm_set1_epi8(2)), _mm_set1_epi8(33));
-    __m128i has_backslash = _mm_cmpeq_epi8(chunk, _mm_set1_epi8('\\'));
-    __m128i needs_escape  = _mm_or_si128(too_low_or_dbl_quote, has_backslash);
-    return _mm_movemask_epi8(needs_escape);
-}
-
 static inline TARGET_SSE2 FORCE_INLINE unsigned char search_escape_basic_sse2(search_state *search)
 {
     if (RB_UNLIKELY(search->has_matches)) {
@@ -476,7 +440,7 @@ static inline TARGET_SSE2 FORCE_INLINE unsigned char search_escape_basic_sse2(se
         if (search->matches_mask > 0) {
             return sse2_next_match(search);
         } else {
-            // sse2_next_match will only advance search->ptr up to the last matching character. 
+            // sse2_next_match will only advance search->ptr up to the last matching character.
             // Skip over any characters in the last chunk that occur after the last match.
             search->has_matches = false;
             if (RB_UNLIKELY(search->chunk_base + sizeof(__m128i) >= search->end)) {
@@ -487,29 +451,22 @@ static inline TARGET_SSE2 FORCE_INLINE unsigned char search_escape_basic_sse2(se
         }
     }
 
-    while (search->ptr + sizeof(__m128i) <= search->end) {
-        int needs_escape_mask = sse2_update(search->ptr);
-
-        if (needs_escape_mask == 0) {
-            search->ptr += sizeof(__m128i);
-            continue;
-        }
-
+    if (string_scan_simd_sse2(&search->ptr, search->end, &search->matches_mask)) {
         search->has_matches = true;
-        search->matches_mask = needs_escape_mask;
         search->chunk_base = search->ptr;
+        search->chunk_end = search->ptr + sizeof(__m128i);
         return sse2_next_match(search);
     }
 
-    // There are fewer than 16 bytes left. 
+    // There are fewer than 16 bytes left.
     unsigned long remaining = (search->end - search->ptr);
     if (remaining >= SIMD_MINIMUM_THRESHOLD) {
         char *s = copy_remaining_bytes(search, sizeof(__m128i), remaining);
 
-        int needs_escape_mask = sse2_update(s);
+        int needs_escape_mask = compute_chunk_mask_sse2(s);
 
         if (needs_escape_mask == 0) {
-            // Nothing to escape, ensure search_flush doesn't do anything by setting 
+            // Nothing to escape, ensure search_flush doesn't do anything by setting
             // search->cursor to search->ptr.
             fbuffer_consumed(search->buffer, remaining);
             search->ptr = search->end;
@@ -638,7 +595,8 @@ static inline unsigned char search_ascii_only_escape(search_state *search, const
     return 0;
 }
 
-static inline void full_escape_UTF8_char(search_state *search, unsigned char ch_len) {
+static inline void full_escape_UTF8_char(search_state *search, unsigned char ch_len)
+{
     const unsigned char ch = (unsigned char)*search->ptr;
     switch (ch_len) {
         case 1: {
@@ -668,7 +626,7 @@ static inline void full_escape_UTF8_char(search_state *search, unsigned char ch_
 
             uint32_t wchar = 0;
 
-            switch(ch_len) {
+            switch (ch_len) {
                 case 2:
                     wchar = ch & 0x1F;
                     break;
@@ -828,7 +786,8 @@ static VALUE mHash_to_json(int argc, VALUE *argv, VALUE self)
  * _state_ is a JSON::State object, that can also be used to configure the
  * produced JSON string output further.
  */
-static VALUE mArray_to_json(int argc, VALUE *argv, VALUE self) {
+static VALUE mArray_to_json(int argc, VALUE *argv, VALUE self)
+{
     rb_check_arity(argc, 0, 1);
     VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
     return cState_partial_generate(Vstate, self, generate_json_array, Qfalse);
@@ -885,17 +844,6 @@ static VALUE mFloat_to_json(int argc, VALUE *argv, VALUE self)
     return cState_partial_generate(Vstate, self, generate_json_float, Qfalse);
 }
 
-/*
- * call-seq: String.included(modul)
- *
- * Extends _modul_ with the String::Extend module.
- */
-static VALUE mString_included_s(VALUE self, VALUE modul) {
-    VALUE result = rb_funcall(modul, i_extend, 1, mString_Extend);
-    rb_call_super(1, &modul);
-    return result;
-}
-
 /*
  * call-seq: to_json(*)
  *
@@ -910,51 +858,6 @@ static VALUE mString_to_json(int argc, VALUE *argv, VALUE self)
     return cState_partial_generate(Vstate, self, generate_json_string, Qfalse);
 }
 
-/*
- * call-seq: to_json_raw_object()
- *
- * This method creates a raw object hash, that can be nested into
- * other data structures and will be generated as a raw string. This
- * method should be used, if you want to convert raw strings to JSON
- * instead of UTF-8 strings, e. g. binary data.
- */
-static VALUE mString_to_json_raw_object(VALUE self)
-{
-    VALUE ary;
-    VALUE result = rb_hash_new();
-    rb_hash_aset(result, rb_funcall(mJSON, i_create_id, 0), rb_class_name(rb_obj_class(self)));
-    ary = rb_funcall(self, i_unpack, 1, rb_str_new2("C*"));
-    rb_hash_aset(result, rb_utf8_str_new_lit("raw"), ary);
-    return result;
-}
-
-/*
- * call-seq: to_json_raw(*args)
- *
- * This method creates a JSON text from the result of a call to
- * to_json_raw_object of this String.
- */
-static VALUE mString_to_json_raw(int argc, VALUE *argv, VALUE self)
-{
-    VALUE obj = mString_to_json_raw_object(self);
-    Check_Type(obj, T_HASH);
-    return mHash_to_json(argc, argv, obj);
-}
-
-/*
- * call-seq: json_create(o)
- *
- * Raw Strings are JSON Objects (the raw bytes are stored in an array for the
- * key "raw"). The Ruby String can be created by this module method.
- */
-static VALUE mString_Extend_json_create(VALUE self, VALUE o)
-{
-    VALUE ary;
-    Check_Type(o, T_HASH);
-    ary = rb_hash_aref(o, rb_str_new2("raw"));
-    return rb_funcall(ary, i_pack, 1, rb_str_new2("C*"));
-}
-
 /*
  * call-seq: to_json(*)
  *
@@ -1093,8 +996,11 @@ static inline VALUE vstate_get(struct generate_json_data *data)
 }
 
 struct hash_foreach_arg {
+    VALUE hash;
     struct generate_json_data *data;
-    int iter;
+    int first_key_type;
+    bool first;
+    bool mixed_keys_encountered;
 };
 
 static VALUE
@@ -1112,6 +1018,29 @@ convert_string_subclass(VALUE key)
     return key_to_s;
 }
 
+NOINLINE()
+static void
+json_inspect_hash_with_mixed_keys(struct hash_foreach_arg *arg)
+{
+    if (arg->mixed_keys_encountered) {
+        return;
+    }
+    arg->mixed_keys_encountered = true;
+
+    JSON_Generator_State *state = arg->data->state;
+    if (state->on_duplicate_key != JSON_IGNORE) {
+        VALUE do_raise = state->on_duplicate_key == JSON_RAISE ? Qtrue : Qfalse;
+        rb_funcall(mJSON, rb_intern("on_mixed_keys_hash"), 2, arg->hash, do_raise);
+    }
+}
+
+static VALUE
+json_call_as_json(JSON_Generator_State *state, VALUE object, VALUE is_key)
+{
+    VALUE proc_args[2] = {object, is_key};
+    return rb_proc_call_with_block(state->as_json, 2, proc_args, Qnil);
+}
+
 static int
 json_object_i(VALUE key, VALUE val, VALUE _arg)
 {
@@ -1122,21 +1051,33 @@ json_object_i(VALUE key, VALUE val, VALUE _arg)
     JSON_Generator_State *state = data->state;
 
     long depth = state->depth;
-    int j;
+    int key_type = rb_type(key);
+
+    if (arg->first) {
+        arg->first = false;
+        arg->first_key_type = key_type;
+    }
+    else {
+        fbuffer_append_char(buffer, ',');
+    }
 
-    if (arg->iter > 0) fbuffer_append_char(buffer, ',');
     if (RB_UNLIKELY(data->state->object_nl)) {
         fbuffer_append_str(buffer, data->state->object_nl);
     }
     if (RB_UNLIKELY(data->state->indent)) {
-        for (j = 0; j < depth; j++) {
-            fbuffer_append_str(buffer, data->state->indent);
-        }
+        fbuffer_append_str_repeat(buffer, data->state->indent, depth);
     }
 
     VALUE key_to_s;
-    switch(rb_type(key)) {
+    bool as_json_called = false;
+
+  start:
+    switch (key_type) {
         case T_STRING:
+            if (RB_UNLIKELY(arg->first_key_type != T_STRING)) {
+                json_inspect_hash_with_mixed_keys(arg);
+            }
+
             if (RB_LIKELY(RBASIC_CLASS(key) == rb_cString)) {
                 key_to_s = key;
             } else {
@@ -1144,9 +1085,23 @@ json_object_i(VALUE key, VALUE val, VALUE _arg)
             }
             break;
         case T_SYMBOL:
+            if (RB_UNLIKELY(arg->first_key_type != T_SYMBOL)) {
+                json_inspect_hash_with_mixed_keys(arg);
+            }
+
             key_to_s = rb_sym2str(key);
             break;
         default:
+            if (data->state->strict) {
+                if (RTEST(data->state->as_json) && !as_json_called) {
+                    key = json_call_as_json(data->state, key, Qtrue);
+                    key_type = rb_type(key);
+                    as_json_called = true;
+                    goto start;
+                } else {
+                    raise_generator_error(key, "%"PRIsVALUE" not allowed as object key in JSON", CLASS_OF(key));
+                }
+            }
             key_to_s = rb_convert_type(key, T_STRING, "String", "to_s");
             break;
     }
@@ -1161,7 +1116,6 @@ json_object_i(VALUE key, VALUE val, VALUE _arg)
     if (RB_UNLIKELY(state->space)) fbuffer_append_str(buffer, data->state->space);
     generate_json(buffer, data, val);
 
-    arg->iter++;
     return ST_CONTINUE;
 }
 
@@ -1177,7 +1131,6 @@ static inline long increase_depth(struct generate_json_data *data)
 
 static void generate_json_object(FBuffer *buffer, struct generate_json_data *data, VALUE obj)
 {
-    int j;
     long depth = increase_depth(data);
 
     if (RHASH_SIZE(obj) == 0) {
@@ -1189,8 +1142,9 @@ static void generate_json_object(FBuffer *buffer, struct generate_json_data *dat
     fbuffer_append_char(buffer, '{');
 
     struct hash_foreach_arg arg = {
+        .hash = obj,
         .data = data,
-        .iter = 0,
+        .first = true,
     };
     rb_hash_foreach(obj, json_object_i, (VALUE)&arg);
 
@@ -1198,9 +1152,7 @@ static void generate_json_object(FBuffer *buffer, struct generate_json_data *dat
     if (RB_UNLIKELY(data->state->object_nl)) {
         fbuffer_append_str(buffer, data->state->object_nl);
         if (RB_UNLIKELY(data->state->indent)) {
-            for (j = 0; j < depth; j++) {
-                fbuffer_append_str(buffer, data->state->indent);
-            }
+            fbuffer_append_str_repeat(buffer, data->state->indent, depth);
         }
     }
     fbuffer_append_char(buffer, '}');
@@ -1208,7 +1160,6 @@ static void generate_json_object(FBuffer *buffer, struct generate_json_data *dat
 
 static void generate_json_array(FBuffer *buffer, struct generate_json_data *data, VALUE obj)
 {
-    int i, j;
     long depth = increase_depth(data);
 
     if (RARRAY_LEN(obj) == 0) {
@@ -1219,15 +1170,13 @@ static void generate_json_array(FBuffer *buffer, struct generate_json_data *data
 
     fbuffer_append_char(buffer, '[');
     if (RB_UNLIKELY(data->state->array_nl)) fbuffer_append_str(buffer, data->state->array_nl);
-    for(i = 0; i < RARRAY_LEN(obj); i++) {
+    for (int i = 0; i < RARRAY_LEN(obj); i++) {
         if (i > 0) {
             fbuffer_append_char(buffer, ',');
             if (RB_UNLIKELY(data->state->array_nl)) fbuffer_append_str(buffer, data->state->array_nl);
         }
         if (RB_UNLIKELY(data->state->indent)) {
-            for (j = 0; j < depth; j++) {
-                fbuffer_append_str(buffer, data->state->indent);
-            }
+            fbuffer_append_str_repeat(buffer, data->state->indent, depth);
         }
         generate_json(buffer, data, RARRAY_AREF(obj, i));
     }
@@ -1235,9 +1184,7 @@ static void generate_json_array(FBuffer *buffer, struct generate_json_data *data
     if (RB_UNLIKELY(data->state->array_nl)) {
         fbuffer_append_str(buffer, data->state->array_nl);
         if (RB_UNLIKELY(data->state->indent)) {
-            for (j = 0; j < depth; j++) {
-                fbuffer_append_str(buffer, data->state->indent);
-            }
+            fbuffer_append_str_repeat(buffer, data->state->indent, depth);
         }
     }
     fbuffer_append_char(buffer, ']');
@@ -1304,7 +1251,7 @@ static void generate_json_string(FBuffer *buffer, struct generate_json_data *dat
     search.chunk_base = NULL;
 #endif /* HAVE_SIMD */
 
-    switch(rb_enc_str_coderange(obj)) {
+    switch (rb_enc_str_coderange(obj)) {
         case ENC_CODERANGE_7BIT:
         case ENC_CODERANGE_VALID:
             if (RB_UNLIKELY(data->state->ascii_only)) {
@@ -1389,7 +1336,7 @@ static void generate_json_float(FBuffer *buffer, struct generate_json_data *data
         /* for NaN and Infinity values we either raise an error or rely on Float#to_s. */
         if (!allow_nan) {
             if (data->state->strict && data->state->as_json) {
-                VALUE casted_obj = rb_proc_call_with_block(data->state->as_json, 1, &obj, Qnil);
+                VALUE casted_obj = json_call_as_json(data->state, obj, Qfalse);
                 if (casted_obj != obj) {
                     increase_depth(data);
                     generate_json(buffer, data, casted_obj);
@@ -1406,12 +1353,11 @@ static void generate_json_float(FBuffer *buffer, struct generate_json_data *data
     }
 
     /* This implementation writes directly into the buffer. We reserve
-     * the 28 characters that fpconv_dtoa states as its maximum.
+     * the 32 characters that fpconv_dtoa states as its maximum.
      */
-    fbuffer_inc_capa(buffer, 28);
+    fbuffer_inc_capa(buffer, 32);
     char* d = buffer->ptr + buffer->len;
     int len = fpconv_dtoa(value, d);
-
     /* fpconv_dtoa converts a float to its shortest string representation,
      * but it adds a ".0" if this is a plain integer.
      */
@@ -1478,7 +1424,7 @@ start:
             general:
                 if (data->state->strict) {
                     if (RTEST(data->state->as_json) && !as_json_called) {
-                        obj = rb_proc_call_with_block(data->state->as_json, 1, &obj, Qnil);
+                        obj = json_call_as_json(data->state, obj, Qfalse);
                         as_json_called = true;
                         goto start;
                     } else {
@@ -1899,6 +1845,19 @@ static VALUE cState_ascii_only_set(VALUE self, VALUE enable)
     return Qnil;
 }
 
+static VALUE cState_allow_duplicate_key_p(VALUE self)
+{
+    GET_STATE(self);
+    switch (state->on_duplicate_key) {
+        case JSON_IGNORE:
+            return Qtrue;
+        case JSON_DEPRECATED:
+            return Qnil;
+        default:
+            return Qfalse;
+    }
+}
+
 /*
  * call-seq: depth
  *
@@ -1956,15 +1915,30 @@ static VALUE cState_buffer_initial_length_set(VALUE self, VALUE buffer_initial_l
     return Qnil;
 }
 
+struct configure_state_data {
+    JSON_Generator_State *state;
+    VALUE vstate;  // Ruby object that owns the state, or Qfalse if stack-allocated
+};
+
+static inline void state_write_value(struct configure_state_data *data, VALUE *field, VALUE value)
+{
+    if (RTEST(data->vstate)) {
+        RB_OBJ_WRITE(data->vstate, field, value);
+    } else {
+        *field = value;
+    }
+}
+
 static int configure_state_i(VALUE key, VALUE val, VALUE _arg)
 {
-    JSON_Generator_State *state = (JSON_Generator_State *)_arg;
+    struct configure_state_data *data = (struct configure_state_data *)_arg;
+    JSON_Generator_State *state = data->state;
 
-         if (key == sym_indent)                { state->indent = string_config(val); }
-    else if (key == sym_space)                 { state->space = string_config(val); }
-    else if (key == sym_space_before)          { state->space_before = string_config(val); }
-    else if (key == sym_object_nl)             { state->object_nl = string_config(val); }
-    else if (key == sym_array_nl)              { state->array_nl = string_config(val); }
+         if (key == sym_indent)                { state_write_value(data, &state->indent, string_config(val)); }
+    else if (key == sym_space)                 { state_write_value(data, &state->space, string_config(val)); }
+    else if (key == sym_space_before)          { state_write_value(data, &state->space_before, string_config(val)); }
+    else if (key == sym_object_nl)             { state_write_value(data, &state->object_nl, string_config(val)); }
+    else if (key == sym_array_nl)              { state_write_value(data, &state->array_nl, string_config(val)); }
     else if (key == sym_max_nesting)           { state->max_nesting = long_config(val); }
     else if (key == sym_allow_nan)             { state->allow_nan = RTEST(val); }
     else if (key == sym_ascii_only)            { state->ascii_only = RTEST(val); }
@@ -1973,11 +1947,16 @@ static int configure_state_i(VALUE key, VALUE val, VALUE _arg)
     else if (key == sym_script_safe)           { state->script_safe = RTEST(val); }
     else if (key == sym_escape_slash)          { state->script_safe = RTEST(val); }
     else if (key == sym_strict)                { state->strict = RTEST(val); }
-    else if (key == sym_as_json)               { state->as_json = RTEST(val) ? rb_convert_type(val, T_DATA, "Proc", "to_proc") : Qfalse; }
+    else if (key == sym_allow_duplicate_key)   { state->on_duplicate_key = RTEST(val) ? JSON_IGNORE : JSON_RAISE; }
+    else if (key == sym_as_json)               {
+        VALUE proc = RTEST(val) ? rb_convert_type(val, T_DATA, "Proc", "to_proc") : Qfalse;
+        state->as_json_single_arg = proc && rb_proc_arity(proc) == 1;
+        state_write_value(data, &state->as_json, proc);
+    }
     return ST_CONTINUE;
 }
 
-static void configure_state(JSON_Generator_State *state, VALUE config)
+static void configure_state(JSON_Generator_State *state, VALUE vstate, VALUE config)
 {
     if (!RTEST(config)) return;
 
@@ -1985,15 +1964,20 @@ static void configure_state(JSON_Generator_State *state, VALUE config)
 
     if (!RHASH_SIZE(config)) return;
 
+    struct configure_state_data data = {
+        .state = state,
+        .vstate = vstate
+    };
+
     // We assume in most cases few keys are set so it's faster to go over
     // the provided keys than to check all possible keys.
-    rb_hash_foreach(config, configure_state_i, (VALUE)state);
+    rb_hash_foreach(config, configure_state_i, (VALUE)&data);
 }
 
 static VALUE cState_configure(VALUE self, VALUE opts)
 {
     GET_STATE(self);
-    configure_state(state, opts);
+    configure_state(state, self, opts);
     return self;
 }
 
@@ -2001,7 +1985,7 @@ static VALUE cState_m_generate(VALUE klass, VALUE obj, VALUE opts, VALUE io)
 {
     JSON_Generator_State state = {0};
     state_init(&state);
-    configure_state(&state, opts);
+    configure_state(&state, Qfalse, opts);
 
     char stack_buffer[FBUFFER_STACK_SIZE];
     FBuffer buffer = {
@@ -2090,6 +2074,8 @@ void Init_generator(void)
     rb_define_method(cState, "generate", cState_generate, -1);
     rb_define_alias(cState, "generate_new", "generate"); // :nodoc:
 
+    rb_define_private_method(cState, "allow_duplicate_key?", cState_allow_duplicate_key_p, 0);
+
     rb_define_singleton_method(cState, "generate", cState_m_generate, 3);
 
     VALUE mGeneratorMethods = rb_define_module_under(mGenerator, "GeneratorMethods");
@@ -2117,13 +2103,7 @@ void Init_generator(void)
     rb_define_method(mFloat, "to_json", mFloat_to_json, -1);
 
     VALUE mString = rb_define_module_under(mGeneratorMethods, "String");
-    rb_define_singleton_method(mString, "included", mString_included_s, 1);
     rb_define_method(mString, "to_json", mString_to_json, -1);
-    rb_define_method(mString, "to_json_raw", mString_to_json_raw, -1);
-    rb_define_method(mString, "to_json_raw_object", mString_to_json_raw_object, 0);
-
-    mString_Extend = rb_define_module_under(mString, "Extend");
-    rb_define_method(mString_Extend, "json_create", mString_Extend_json_create, 1);
 
     VALUE mTrueClass = rb_define_module_under(mGeneratorMethods, "TrueClass");
     rb_define_method(mTrueClass, "to_json", mTrueClass_to_json, -1);
@@ -2160,6 +2140,7 @@ void Init_generator(void)
     sym_escape_slash = ID2SYM(rb_intern("escape_slash"));
     sym_strict = ID2SYM(rb_intern("strict"));
     sym_as_json = ID2SYM(rb_intern("as_json"));
+    sym_allow_duplicate_key = ID2SYM(rb_intern("allow_duplicate_key"));
 
     usascii_encindex = rb_usascii_encindex();
     utf8_encindex = rb_utf8_encindex();
@@ -2168,7 +2149,7 @@ void Init_generator(void)
     rb_require("json/ext/generator/state");
 
 
-    switch(find_simd_implementation()) {
+    switch (find_simd_implementation()) {
 #ifdef HAVE_SIMD
 #ifdef HAVE_SIMD_NEON
         case SIMD_NEON: